Method and system for automatically generating energy policy brief report
By collecting and preprocessing energy policy data, using generator and discriminator models to generate and optimize briefings, and through real-time tracking and user feedback mechanisms, the time-consuming and labor-intensive generation of traditional energy policy briefings is solved, achieving efficient and personalized energy policy briefings.
Patent Information
- Application Number
- CN202411918959.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-25
- Publication Date
- 2025-05-30
AI Technical Summary
Traditional energy policy briefing generation methods are time-consuming and labor-intensive, making it difficult to ensure the quality and personalized needs of the newsletter generation, and there is a lot of workload and error risk when manually processing and analyzing energy policy information.
By collecting and preprocessing energy policy-related data, using generator and discriminator models for briefing generation and optimization, setting up real-time tracking and dynamic update mechanisms, adding user feedback mechanisms, and generating a mesh knowledge graph through natural language processing technology.
It has achieved automated generation of energy policy briefs, improved the efficiency and reading quality of information updates, met the personalized needs of different users, and provided high-quality and personalized energy policy briefs.
Smart Images
Figure CN120067287A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of energy policies, and particularly to a method and system for automatically generating energy policy briefs. Background Art
[0002] In the traditional energy policy release and analysis process, people usually need to read a large number of policy documents to understand the specific content of new policies on the one hand and analyze and predict the impact of new policies on the energy market on the other hand. This is a time-consuming and laborious task.
[0003] Currently, government departments and enterprises and institutions around the world release a large amount of energy policy-related information every day. The workload of manually collecting and processing this information is very large and error-prone. For government departments and enterprises and institutions, it is necessary to sort out and summarize the collected energy policy information to generate corresponding policy briefs for the use of decision-makers and relevant personnel. This process usually requires professional writing skills and a lot of time, and it is difficult to ensure the quality of the generated briefs. Moreover, different users may have different concerns and needs regarding energy policies, so the generated briefs will also vary. However, traditional brief generation methods usually have difficulty meeting such personalized needs. Summary of the Invention
[0004] The purpose of this part is to outline some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this part, as well as in the abstract and title of the specification of this application, to avoid obscuring the purpose of this part, the abstract, and the title. However, such simplifications or omissions shall not be used to limit the scope of the present invention.
[0005] In view of the problems existing in the above method for automatically generating energy policy briefs, the present invention is proposed.
[0006] To solve the above technical problems, the present invention provides the following technical solutions: collecting data and preprocessing the data to convert it into a numerical form; inputting the numerical data into a generator and a discriminator model to generate and optimize a brief; setting up real-time tracking and dynamic updating and adding a user feedback mechanism to optimize the content of the brief again; and converting the optimized content of the brief into a network knowledge graph through natural language processing technology.
[0007] As a preferred solution of the method for automatically generating energy policy briefs according to the present invention, the method for collecting data includes setting multi-language crawling rules and collecting information from energy policy data sources in different languages through an internationalized crawler framework.
[0008] As a preferred solution of the automatic generation method of the energy policy briefing described in the present invention, wherein: the preprocessing includes operations such as cleaning, standardizing, word segmentation, stop word removal, stemming, and word embedding on the collected data to form structured data; the word embedding uses the TF-IDF method to convert text data into a numerical form, and the calculation formula is as follows:
[0009] TF-IDF(t,d) = TF(t,d) * IDF(t)
[0010] IDF(t) = log(n / (1 + df(t)))
[0011] Wherein, TF(t,d) represents the frequency of occurrence of word t in document d, and IDF(t) represents the inverse document frequency of word t, n represents the total number of documents, and df(t) represents the number of documents containing word t.
[0012] As a preferred solution of the automatic generation method of the energy policy briefing described in the present invention, wherein: the generator is used to generate the energy policy briefing to be discriminated, so that the discriminator cannot distinguish the difference between the briefing generated by it and the real briefing. The loss function of the generator is defined as:
[0013] LG = -E X～pd[X] log(D(G(x)))
[0014] Wherein, pd[X] is the distribution of the real policy briefing, G(x) is the briefing generated by the generator, D(G(x)) represents the probability that the discriminator judges the generated briefing as a real briefing, and E is the mathematical expectation.
[0015] As a preferred solution of the automatic generation method of the energy policy briefing described in the present invention, wherein: the discriminator is used to correctly identify the real briefing and the generated briefing. The loss function of the discriminator D is defined as:
[0016] LD = -E x～pd[x] [logD(x)] - E x～pg[x] log(1 - D(G(x)))
[0017] Wherein, pg[X] is the distribution of the briefing generated by the generator, and D(x) represents the probability that the discriminator judges the input x as a real briefing.
[0018] As a preferred solution of the automatic generation method of the energy policy briefing described in the present invention, wherein: the generator model and the discriminator model carry out cyclic training. The method of the cyclic training includes fixing the generator and training the discriminator with real data and the data generated by the generator; fixing the discriminator and training the generator until the generator and the discriminator reach the Nash equilibrium.
[0019] As a preferred solution of the energy policy briefing automatic generation method of the present invention, wherein: the real-time tracking and dynamic update include setting a timed crawler task to regularly collect new policy data from a specified data source and adding the latest data in real time when the model generates a briefing.
[0020] As a preferred solution of the energy policy briefing automatic generation method of the present invention, wherein: the user feedback mechanism adjusts the model parameters or generation strategy by recording the user's evaluation of the generated briefing, including accuracy, content integrity, and personalized needs.
[0021] As a preferred solution of the energy policy briefing automatic generation method of the present invention, wherein: the knowledge graph construction step includes extracting named entities and their relationships in the policy text; constructing nodes and edges based on the extracted entities and relationships; and dynamically expanding the knowledge graph through an incremental update mechanism.
[0022] To solve the above technical problems, the present invention also provides the following technical solution: A system for the energy policy briefing automatic generation method adopts the energy policy briefing automatic generation method, including a preprocessing module, a model construction and training module, a real-time update module, a user feedback module, and a knowledge graph module; the preprocessing module is used to collect text data from energy policy-related data sources around the world and perform cleaning, word segmentation, stop word removal, stemming, and word embedding processing on the collected text data; the model construction and training module designs a generator and a discriminator based on a pre-trained model combined with a generative adversarial network and continuously optimizes the quality of the generated briefing through the adversarial learning of the generator and the discriminator; the real-time update module updates the briefing content through real-time data collection and model fine-tuning; the user feedback module dynamically adjusts the generation strategy according to user feedback; the knowledge graph module generates an energy policy networked knowledge graph through semantic abstraction technology.
[0023] The beneficial effects of the present invention are as follows: The present invention can collect text data related to energy policies from speeches and policy announcements around the world, complete comprehensive policy collection and analysis. Through real-time tracking and dynamic update, the timeliness and accuracy of the information in the policy report are ensured. Using deep learning technology, it can automatically generate energy policy briefings with high quality and high authenticity. By adding a user feedback mechanism, it can adjust the content generation strategy according to user feedback, achieve personalized dynamic adjustment, and meet the reading needs of different users. Applying natural language processing technology to semantically abstract the generated briefing to generate a networked knowledge graph, enabling readers to obtain and understand policy information more conveniently. In short, the present invention realizes the automatic generation of energy policy briefings, not only improving the efficiency of information update, but also improving the reading quality and user experience, providing important help for the decision-making of the public, enterprises, policy makers, etc. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.
[0025] Figure 1 It is a flowchart of an automatic generation method for energy policy briefs. Specific Embodiments
[0026] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the specific embodiments of the present invention in detail with reference to the drawings of the specification.
[0027] In the following description, many specific details are set forth to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.
[0028] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it an embodiment that is separately or selectively mutually exclusive with other embodiments.
[0029] Embodiment 1
[0030] Referring to Figure 1 , this is the first embodiment of the present invention, which provides an automatic generation method for energy policy briefs.
[0031] S1. Collect data, and collect text data such as news and policy announcements related to energy policies from all over the world;
[0032] Step S1 is specifically as follows:
[0033] S11. Define the data source and determine the sources for collecting policy data, such as government official websites, news websites, social media platforms, professional energy websites, subscribed databases, etc.
[0034] S12. Develop a collection strategy, and develop a suitable data collection strategy according to the characteristics of the data source and the required data. For example, for different government official websites, different crawler strategies may need to be designed.
[0035] S13. Write a crawler program using Python or other programming languages, locate specific data according to the formulated strategy, and extract the required information. For example, extract news titles, content, release time, etc. related to energy policies from news websites. Set multi-language crawling rules, and collect data from energy policy websites around the world through an internationalized crawler framework such as Scrapy.
[0036] S2. Data preprocessing, clean and process the collected data to prepare for subsequent model training;
[0037] Step S2 is specifically as follows:
[0038] S21. Data cleaning, remove invalid, incomplete, incorrect or irrelevant records. Remove HTML tags from news articles, remove articles not related to energy policies, and clear obvious typos or garbled codes, etc. Data standardization, convert all text to lowercase letters to reduce data complexity. Then divide the text data into words or phrases using tokenization tools such as NLTK and jieba. Remove some common words with no actual meaning, such as the prepositions "and" and "the". Perform stemming and lemmatization. For example, "running", "runs" and "ran" will all be reduced to "run". This can reduce the size of the vocabulary and enable the model to generalize better.
[0039] S22. Word embedding, convert text into a numerical form for mathematical calculations, using the TF-IDF method, and the calculation formula is as follows:
[0040] TF-IDF(t,d) = TF(t,d) * IDF(t)
[0041] IDF(t) = log(n / (1 + df(t)))
[0042] Among them, TF(t,d) represents the frequency of occurrence of word t in document d, while IDF(t) represents the inverse document frequency of word t, n represents the total number of documents, and df(t) represents the number of documents containing word t. In actual use, to prevent the denominator from being zero, usually 1 is added to the denominator.
[0043] Embodiment 2
[0044] Refer to Figure 1 , which is the second embodiment of the present invention. This embodiment further illustrates how an energy policy briefing is automatically generated based on the previous embodiment.
[0045] S3. Build a deep learning model, based on a pre-trained model (BERT), combined with a generative adversarial network (GAN);
[0046] First, the BERT model is adopted. It is a pre-trained model based on Transformer that can effectively learn the semantics and context in the text and is suitable for processing various NLP tasks. At the same time, the generative adversarial network (GAN) is combined. The generator is responsible for generating energy policy briefs, and the discriminator is responsible for judging whether the generated briefs are real and effective. In the present invention, the BERT model is adjusted and fine-tuned on energy policy data to adapt to the text generation task of the briefs. The loss function of the model is usually the negative log-likelihood function:
[0047] L = -ΣylogP(y|x)
[0048] where x is the input, y is the prediction, and P(y|x) represents the probability of predicting y given the input x.
[0049] The generative adversarial network (GAN) includes a generator G and a discriminator D. The generator attempts to generate new data that looks like real data, and the task of the discriminator is to distinguish real data from real and generated data. The generator G and the discriminator D can use the BERT model as the infrastructure. The goal of the generator G is to generate energy policy briefs that look real, so that the discriminator D cannot distinguish the difference between the briefs it generates and real briefs. The loss function of the generator G is defined as:
[0050] LG = -E X～pd[X] log(D(G(x)))
[0051] where pd[X] is the distribution of real policy briefs, G(x) is the brief generated by the generator, D(G(x)) represents the probability that the discriminator judges the generated brief as a real brief, and E is the mathematical expectation.
[0052] The goal of the discriminator D is to correctly identify real briefs and generated briefs. The loss function of the discriminator D is defined as:
[0053] LD = -E x～pd[x] [logD(x)] - E x～pg[x] log(1 - D(G(x)))
[0054] where pg[X] is the distribution of the briefs generated by the generator, and D(x) represents the probability that the discriminator judges the input x as a real brief.
[0055] S4. Model training: Using the existing policy text data, the model parameters are updated by optimizing the loss function, so that the model learns how to generate high-quality and highly realistic energy policy briefs;
[0056] Step S4 is specifically as follows:
[0057] S41. Data Preparation: Preprocess the existing policy text data, including steps such as word segmentation, building a dictionary, and encoding the text into word vectors.
[0058] S42. Initialize the Model: Fine-tune the pre-trained BERT model, including initializing the model's parameters; initialize the generator G and discriminator D of the GAN.
[0059] S43. Train the Generator and Discriminator: By alternately optimizing the distribution of the generated briefing, each step of training consists of two steps:
[0060] The first step: Fix the generator G and optimize the parameters of the discriminator D. Train the discriminator with real data and the data generated by the generator, and use optimization algorithms such as Stochastic Gradient Descent (SGD) to update the model parameters of the discriminator, so that the discriminator can better identify real briefings and generated briefings.
[0061] The second step: Fix the discriminator D and optimize the parameters of the generator G. Still use the optimization algorithm to update the model parameters of the discriminator, so that the briefing generated by the generator is more like a real briefing, making it more difficult for the discriminator to distinguish between the generated briefing and the real briefing.
[0062] Repeat the above two steps, usually for dozens to hundreds of rounds, until the generator and discriminator reach a Nash equilibrium, that is, a balance is achieved between the generator maximizing its score and the discriminator maximizing its ability to identify real and generated data.
[0063] S44. Model Evaluation: After each round of training, evaluate the model through the test set and adjust hyperparameters such as the learning rate to optimize the model performance.
[0064] Through the adversarial learning of the generator and discriminator, the generator gradually learns how to create more realistic briefings, and the discriminator is also constantly learning how to more accurately distinguish between real and generated briefings.
[0065] Embodiment 3
[0066] Refer to Figure 1 , which is the third embodiment of the present invention. This embodiment is based on the first two embodiments.
[0067] S5. Real-time Tracking and Dynamic Update: Set up a real-time tracking system to collect and analyze new energy policies in real time and update them into the briefing to ensure the timeliness and accuracy of the briefing;
[0068] The specific steps of S5 are as follows:
[0069] S51. Set up a scheduled crawler: Use a scheduled execution script (such as the scheduled task function of the schedule library in Python) or set up a Cron job to enable the data collection program to run at regular intervals, such as once a day or once an hour.
[0070] S52. Data collection and screening: The crawler program will crawl specified news websites, social media, professional forums, etc., collect new data information related to energy policies, and perform preprocessing and screening to filter out new data information that meets the conditions.
[0071] S53. Data update: Update the filtered data to the database so that the model will take into account the latest energy policy information when generating reports.
[0072] S54. Model adjustment: If the new data distribution differs significantly from the old data distribution, it may be necessary to retrain or fine-tune the model to adapt to the new data distribution.
[0073] S6. Add a user feedback mechanism and adjust the content generation strategy according to user feedback to achieve personalized dynamic adjustment. This can help continue to optimize the content of the report and ensure the relevance and accuracy of the information;
[0074] Step S6 is specifically as follows:
[0075] S61. Collect user feedback: Collect user evaluation feedback on the generated report through the system interface. This feedback includes feedback on various aspects such as the report content, report quality, and report accuracy.
[0076] S62. Analyze user feedback: Statistically analyze the collected user feedback to identify parts that need to be improved or optimized in the model.
[0077] S63. Adjust the generation strategy: According to the analysis results, feed the user feedback information back to the model, adjust the content generation strategy or model parameters to optimize the content of the generated report to better meet the user's needs.
[0078] S7. Apply natural language processing (NLP) technology to semantically abstract the generated briefing and generate a network knowledge graph to enable readers to more conveniently obtain and understand policy information;
[0079] Step S7 is specifically as follows:
[0080] S71. Entity extraction, which uses NLP technology to extract key entities from text. In natural language processing, entity extraction often uses Named Entity Recognition (NER) technology to achieve. For example, key entities such as policy names, company names, time, city names, etc. can be extracted through a pre-trained BERT model. A extraction function E(x) can be defined to represent this step:
[0081] E(x) = {e1, e2,..., en}
[0082] Here, x is the input text, and e1, e2,..., en are the extracted entities.
[0083] S72. Relationship extraction. After entity extraction, the relationships between entities need to be extracted. This often uses NLP technologies such as dependency analysis. A extraction function R(x) is defined to represent this step:
[0084] R(x) = {r1, r2,... rm}
[0085] Here, x is the input text, and r1, r2,..., rm are the extracted entity relationships.
[0086] S73. Knowledge graph construction and update. Using the extracted entities as nodes and entity relationships as edges, a knowledge graph is constructed. A function G(E, R) is defined to represent this step:
[0087] G = G(E(x), R(x))
[0088] Here, E(x) and R(x) are the entity set and relationship set respectively, and G is the generated knowledge graph.
[0089] When new policy data is updated, the knowledge graph needs to be updated. At this time, it is achieved by defining an update function U(G, E', R'):
[0090] G' = U(G, E'(x), R'(x))
[0091] Here, E'(x) and R'(x) represent the entities and relationships of the new data, G is the original knowledge graph, and G' is the updated knowledge graph.
[0092] S74. Semantic search. With the knowledge graph, semantic search can be realized. This usually requires a semantic matching algorithm, such as the BERT model. A search function S(q, G) is defined to achieve semantic search:
[0093] r = S(q, G)
[0094] Here, q is the user's query, G is the knowledge graph, and r is the returned search result.
[0095] Each step in the above process requires the use of NLP technology and continuous adjustment and optimization in practice. The generated knowledge graph can help users more intuitively understand policy information and achieve accurate and fast semantic search.
[0096] Example 4
[0097] Refer to Figure 1 , which is the fourth embodiment of the present invention. This embodiment is based on the first three embodiments.
[0098] First, set up the steps of data collection and preprocessing. For example, subscribe to the official websites of the US Department of Energy (https: / / www.energy.gov / ), the energy department of the European Commission (https: / / ec.europa.eu / energy / home_en), the journal Energy Policy (https: / / www.journals.elsevier.com / energy-policy), Twitter accounts related to energy policy, etc. as data sources. Then, design and run a crawler program written in Python to regularly collect information such as news articles, announcements, editorials, tweets, etc. containing energy policies from these websites according to the set strategy. For example, the following news report about a new solar project was collected from the official website of the US Department of Energy:
[0099] "The US Department of Energy announced that it will invest 20 million US dollars to promote the research and development of solar projects to drive faster and more efficient solar deployment. This move is to achieve the goal of 30% of the electricity supply in 2025. The project will focus on three major areas: integrated photovoltaic systems, innovative thin-film solar technologies, and solar system performance models and verification."
[0100] In the data preprocessing stage, use the NLTK library in Python to tokenize the text, and the following tokenization results are obtained:
[0101] "['US', 'Department of Energy', 'announced', 'will', 'invest', '20 million', 'US dollars', 'promote','solar project','research and development', 'to', 'drive', 'faster','more', 'efficient','solar', 'deployment', 'this move', 'is', 'to', '
[0102] The project aiming to achieve the goal of 30% electricity supply in 2025 will focus on three major areas: integrated photovoltaic systems, innovative thin-film solar technologies, and solar system performance models and verification.
[0103] Then, use the TF-IDF vectorizer in the Scikit-learn library of Python to perform TF-IDF conversion on the tokenized results to obtain word vectors: ('United States', 0.05), ('Department of Energy', 0.05), ('announced', 0.15), ('will', 0.1), ('invest', 0.15),....
[0104] In the model building stage, the BERT pre-trained model was used, and the generative adversarial network (GAN) was implemented using Pytorch. Subsequently, by alternately training the discriminator and the generator, the quality of the generated reports was gradually improved. For example, through training, a briefing generated by GAN might be like this:
[0105] "The US Department of Energy recently announced that it will invest 20 million US dollars in the research and development of solar projects in order to promote solar deployment more quickly and efficiently. The goal of this measure is to achieve 30% of electricity supply by 2025. The key areas of the project include integrated photovoltaic systems, innovative thin-film solar technologies, and solar system performance models and verification."
[0106] Finally, in the application stage, collect the feedback from users on these generated briefings. For example, users feedback "hope to know more details about integrated photovoltaic systems", and then adjust and optimize the model according to these feedbacks. At the same time, update the knowledge graph according to the newly crawled data, so that users can obtain the latest intelligence when querying.
[0107] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.
Claims
1. A method for automatically generating an energy policy brief, characterized by: Collect data and pre-process the data to convert it into numerical form; Inputting the numerical data into the generator and discriminator models to generate and optimize the briefing; Set up real-time tracking and dynamic updates and add user feedback mechanisms to further optimize the content of the briefing; Through natural language processing technology, the optimized briefing content is converted into a network knowledge graph.
2. The method for automatically generating an energy policy brief as claimed in claim 1, characterized in that: The method for collecting data includes setting multilingual crawling rules and collecting information from energy policy data sources in different languages through an internationalized crawler framework.
3. The method for automatically generating an energy policy briefing as claimed in claim 2, characterized in that: The preprocessing includes cleaning, standardizing, word segmentation, stop word removal, stem extraction, word embedding and other operations on the collected data to form structured data; The word embedding uses the TF-IDF method to convert text data into numerical form. The calculation formula is as follows: TF-IDF(t,d)=TF(t,d)*IDF(t) IDF(t)=log(n / (1+df(t))) Among them, TF(t,d) represents the frequency of occurrence of word t in document d, IDF(t) represents the inverse document frequency of word t, n represents the total number of documents, and df(t) represents the number of documents containing word t.
4. The method for automatically generating an energy policy brief as claimed in claim 3, characterized in that: The generator is used to generate energy policy briefs to be discriminated, so that the discriminator cannot distinguish the difference between the generated briefs and the real briefs. The loss function of the generator is defined as: LG=-E X～pd[X] log(D(G(x))) Where pd[X] is the distribution of true policy briefs, G(x) is the brief generated by the generator, D(G(x)) represents the probability that the discriminator judges the generated brief to be a true brief, and E is the mathematical expectation.
5. The method for automatically generating an energy policy briefing as claimed in claim 4, characterized in that: The discriminator is used to correctly identify the real briefing and the generated briefing. The loss function of the discriminator D is defined as: LD=-E x～pd[x] [logD(x)]-E x～pg[x] log(1-D(G(x))) Here, pg[X] is the distribution of briefs generated by the generator, and D(x) represents the probability that the discriminator judges that the input x is a true brief.
6. The method for automatically generating an energy policy briefing as claimed in claim 5, characterized in that: The generator model and the discriminator model carry out cyclic training, and the cyclic training method includes: The generator is fixed, and the discriminator is trained using real data and data generated by the generator; The discriminator is fixed and the generator is trained until the generator and the discriminator reach a Nash equilibrium.
7. The method for automatically generating an energy policy briefing as claimed in claim 6, characterized in that: The real-time tracking and dynamic updating include setting scheduled crawler tasks, regularly collecting new policy data from designated data sources, and adding the latest data in real time when the model generates a briefing.
8. The method for automatically generating an energy policy briefing as claimed in claim 7, characterized in that: The user feedback mechanism records the user's evaluation of the generated briefing, including accuracy, content completeness and personalized needs, and adjusts the model parameters or generation strategy accordingly.
9. The method for automatically generating an energy policy briefing as claimed in claim 8, characterized in that: The knowledge graph construction steps include: Extract named entities and their relations in policy text; Build nodes and edges based on the extracted entities and relationships; Dynamically expand the knowledge graph through an incremental update mechanism.
10. A system using the method for automatically generating energy policy briefs according to any one of claims 1 to 9, characterized in that: Including preprocessing module, model building and training module, real-time update module, user feedback module, and knowledge graph module; The preprocessing module is used to collect text data from energy policy-related data sources around the world and perform cleaning, word segmentation, stop word removal, stemming and word embedding processing on the collected text data; The model building training module designs a generator and a discriminator based on a pre-trained model combined with a generative adversarial network and continuously optimizes the quality of the generated briefing through adversarial learning of the generator and the discriminator; The real-time update module updates the briefing content through real-time data collection and model fine-tuning; The user feedback module dynamically adjusts the generation strategy according to user feedback; The knowledge graph module generates an energy policy network knowledge graph through semantic abstraction technology.